Citi Bike Use

Jonathan Campbell and Rachel Smith

12-18-2014

Citi Bike furnishes public data that details every trip taken since the beginning of the program.

Notably, the data has already been processed by citibike. They took out all trips under 60 seconds (under the assumption that they might be false starts or people trying to re-dock a bike to make sure it was secure), trips by staff members moving bikes, and trips from “test” stations.

The data contains the following elements:

• trip duration
• start/end time and date
• start/end station name, ID, latitude/longitude
• bike ID
• user type (customer = 24h/7d pass, subscriber = annual subscriber)
• gender
• year of birth

We're interested in evaluating citibike riders, ridership, and revenue.

We chose June 2014 because it is recent enough to be interesting but the ridership is probably not as distorted by heat as July and August and so is probably a better representation of year-round behavior.

We created variables for the age of riders and the distance between start and end stations (as the crow flies):

table(is.na(june$birth.year), june$usertype=="Subscriber") ##more or less, only subscribers have birthdays in the data
june$age <- 2014-as.numeric(june$birth.year)

june$mile <- gdist(june$start.station.longitude, june$start.station.latitude, june$end.station.longitude, june$end.station.latitude, units="miles")

This plots the age and gender of citibike riders:

histogram <- june[,c("age","gender")]
histogram$gender=mapvalues(histogram$gender, from = c(0,1,2), to = c(NA,"MALE","FEMALE"))
histogram <- na.omit(histogram)
age.hist <- ggplot(histogram, aes(x=age, fill=gender)) + geom_histogram(binwidth=1, alpha=.5, position="identity") #by gender

This plots the age and gender of citibike riders:

A hexplot of age versus distance: improbably aged outliers, hard to see patterns in hot spot:

#first attempt at a hexplot for age v distace
hexplot <- june[,c("mile","age")]
hexplot<-na.omit(hexplot)
agehex1 <- ggplot(hexplot, aes(x=age, y=mile)) + stat_binhex(bins=54)

A hexplot of age versus distance: improbably aged outliers, hard to see patterns in hot spot:

Cutting off the age range at 70:

#Cutting off the age range at 70
hexplot1<-hexplot
hexplot1$age<-ifelse(hexplot1$age>70,NA,hexplot1$age) #age cut off at 70
hexplot1<-na.omit(hexplot1)
agehex2 <- ggplot(hexplot1, aes(x=age, y=mile)) + stat_binhex()

Cutting off the age range at 70:

Too zoom in further, we limit the range to 2 miles and 60 years old:

hexplot2<-hexplot1
hexplot2$mile<-ifelse(hexplot2$mile>2,NA,hexplot2$mile) 
hexplot2$age<-ifelse(hexplot2$age>60,NA,hexplot2$age) #age cut off at 70
hexplot2<-na.omit(hexplot2)
agehex3 <- ggplot(hexplot2, aes(x=age, y=mile)) + stat_binhex(bins=22)

Too zoom in further, we limit the range to 2 miles and 60 years old:

This code plots the frequency of trips started at stations on June 15, a Sunday; and June 16, a Monday:

june15 <- june[june$starttime==as.Date("2014-06-15"),]
june16 <- june[june$starttime==as.Date("2014-06-16"),]

june15map <- qmap(c(lon = -73.986029, lat = 40.721111), zoom=12, color = "bw", legend = "topleft") +
geom_point(data=june15, aes(x=start.station.longitude, y=start.station.latitude), alpha=0.01, col="red")
june15zoom <- qmap(c(lon = -73.986029, lat = 40.721111), zoom=13, color = "bw", legend = "topleft") +
geom_point(data=june15, aes(x=start.station.longitude, y=start.station.latitude), alpha=0.01, col="red")
june16map <- qmap(c(lon = -73.986029, lat = 40.721111), zoom=12, color = "bw", legend = "topleft") +
geom_point(data=june15, aes(x=start.station.longitude, y=start.station.latitude), alpha=0.01, col="blue")
june16zoom <- qmap(c(lon = -73.986029, lat = 40.721111), zoom=13, color = "bw", legend = "topleft") +
geom_point(data=june15, aes(x=start.station.longitude, y=start.station.latitude), alpha=0.01, col="blue")

This code build a plot to show the proportion of bike transactions that are bikes leaving the station (the service moves bikes between stations manually to address this inequality):

df1<-data.frame(table(june$start.station.name))
df2<-data.frame(table(june$end.station.name))

#proportion of bikes leaving/arriving at a station
station<-merge(df1,df2,by.x="Var1",by.y="Var1")   
colnames(station)<-c("name","start","end")
station$perleave<-station$start/(station$start+station$end)
station$perarrive<-station$end/(station$start+station$end) 
stationsunique <- june[!duplicated(june$start.station.name),]
start.proportion <- merge(station, stationsunique, by.x="name", by.y="start.station.name")
start.proportion <- start.proportion[,c("perleave", "start.station.latitude", "start.station.longitude")]

stationleave <- qmap("nyc", zoom=12, color = "bw", legend = "topleft") + 
       geom_point(data=start.proportion, aes(x=start.station.longitude, y=start.station.latitude, color=perleave)) + 
       scale_color_gradient(low="yellow", high="red")

Calculating the fines for each person and summed by day of the week:

charge1<-function(x){
  b<-(x/60)-45
  z <- (b%/%30)
  y<-0
  ifelse(b<0,y<-0,
            ifelse(b<=30,y<-2.5,y<-z*9))
return(y)
}

charge2<-function(x){
  b<-(x/60)
  z <- (b%/%30)
  y<-0
  ifelse(z==0,y<-0,
         ifelse(z==1,y<-4,
                ifelse(z==2,y<-13,y<-13+12*(z-2))))
return(y)}
pay<-matrix(nrow=nrow(june),ncol=1)
for (i in 1:nrow(june)){
  ifelse(june[i,"usertype"]=="Customer",
          pay[i]<-charge2(june[i,"tripduration"]),
          pay[i]<-charge1(june[i,"tripduration"]))}
june$pay<-pay

#money vs time of the week chart
june$day <- weekdays(as.Date(june$starttime))
june$hour <- substr(as.character(june$starttime),12,13)
june$dh <- paste(june$day,june$hour)
juneweek <- aggregate(june$pay, list(time=june$dh), sum)
juneweek <- data.frame(juneweek,c(97:120,1:24,121:144,145:168,73:96,25:48,49:72))

juneweek.s <-june[june$usertype=="Subscriber",]
juneweek.s <- aggregate(juneweek.s$pay, list(time=juneweek.s$dh), sum)
juneweek <- merge(juneweek,juneweek.s,by.x="time",by.y="time")
juneweek.c <-june[june$usertype=="Customer",]
juneweek.c <- aggregate(juneweek.c$pay, list(time=juneweek.c$dh), sum)
juneweek <- merge(juneweek,juneweek.c,by.x="time",by.y="time") 

colnames(juneweek)<-c("period","pay","time", "subscriber", "customer")  

This plot shows the frequency of ridership by hour:

Plot of revenue by subscriber type and day of week: http://embed.plnkr.co/jud4z0Gouoa0UWKpJ0oH/preview